Question
Can the samples be removed from the skill after labeling and training?
Answer
To train the documented skill with all samples, that were provided to the software, it is crucial to keep all the samples inside the skill, removing just obsolete or irrelevant ones.
In case you train on 10 samples, then remove them and add another 10 to train on them, only the latest would be applied to the skill and previous training would be considered obsolete with the samples removed.
Comments
2 comments
Richard King
Could you explain why the training versions are not stored in the neuro network and then amended with each new version of trained docs? It seems like a lot of overhead to rebuild the network after each training. That might explain why training takes so long....
Anton Vakhtel
Richard King Hi,
Training the skills incrementally would mean that the latest added documents will take priority over the already existing ones. The only way to make sure that the training is providing better results is to build a new model from scratch and then compare the results with the old one (which is done in the background). As such, it is an architectural decision to retrain the model from scratch and not incrementally.
Best regards,
Anton
Please sign in to leave a comment.